The era of corporate AI theater is ending. A recent Wharton Human-AI Research and GBK Collective study finds that 82% of enterprise leaders use generative AI at least weekly, and 46% use it daily. That is no longer experimentation.

Yet McKinsey’s 2025 state of AI survey shows a stubborn gap between broad usage and real scale, with many companies still stuck in pilots and isolated use cases. The next corporate divide won’t be between firms that “have AI” and firms that don’t. It will be between firms that wire AI into workflows, incentives, and management routines, and firms that keep mistaking access for transformation.

Mainstreaming was the easy part

The report’s most revealing finding isn’t that executives use AI often. It’s where they use it: data analysis, summarization, document creation, and research. Those are high-frequency, repeatable tasks with clear handoffs, which is exactly where the economics of AI are strongest.

Research published in the Quarterly Journal of Economics found that a generative AI assistant raised customer support productivity by 15% on average, with far larger gains for less experienced workers. Meanwhile, Stanford’s 2025 AI Index captures the broader backdrop of fast-improving model capability and expanding economic effect. The lesson for leaders is simple: The first wave of value comes from narrowing work, not mythologizing it.

That’s why adoption gaps now look less like a technology problem and more like a management problem. The Wharton findings show that IT and procurement lead while sales, marketing, and operations still trail.

Teams move faster when they have clear permissions, usable data, and managers who can define acceptable risk. They stall when every prompt feels like a policy exception. The companies pulling ahead aren’t the ones with the loudest AI narrative. They’re the ones making AI boring enough to become routine.

ROI has replaced FOMO

The healthiest number in the study may be 72%: the share of leaders formally measuring AI ROI. Two years ago, boardrooms rewarded enthusiasm—perhaps in fear of missing out (FOMO). Now they’re asking which workflows move, which costs fall, and which revenue lines improve.

That shift mirrors McKinsey’s research on organizations rewiring to capture AI value, and it lines up with PwC’s 2025 AI Jobs Barometer, which found that industries best positioned to use AI are seeing much faster productivity and revenue-per-employee growth.

This is also why internal R&D deserves more scrutiny than applause. Custom tools can create durable advantages when they target a real bottleneck, a proprietary dataset, or a regulated process. They destroy value when they become executive vanity projects. The winners in 2026 will not be the firms with the most co-pilots or the most agent talk. They’ll be the ones that show faster cycle times, better decisions, fewer errors, and cleaner unit economics.

AI has entered its spreadsheet era, and that’s good news.

Talent is now the constraint

If use is mainstream and ROI is measurable, why are so many companies still underperforming?

Because people now set the ceiling. The Wharton study shows stronger C-suite ownership and tighter guardrails, yet it also finds slipping confidence in training and persistent concern about skill erosion. That tension is exactly what NIST’s Generative AI Profile was built to address: Scaling AI safely requires governance that lives inside daily operations, not a compliance memo bolted on at the end.

The labor market is sending the same signal. The World Economic Forum’s Future of Jobs Report 2025 identifies AI and big-data capabilities among the fastest-growing skill demands of the next decade, while an Organisation for Economic Co-operation and Development brief on the AI skills gap warns that training supply might not be enough even for broad AI literacy.

At the same time, PwC reports that workers with AI skills command a substantial wage premium. Companies that cut training while complaining about talent scarcity are effectively bidding against themselves. They will pay more for outside talent while getting less from the people they already employ.

Conclusion

Enterprise AI has entered a more serious phase. The novelty is fading, the budgets are real, and the excuses are running out. What matters now is disciplined workflow design, management accountability, and workforce capability.

Firms that win the next round will treat AI as a system of work, not a collection of tools. Everyone else will keep buying intelligence without ever quite learning how to use it.